Aug 2, 2026

Reinforcing the USB-C and Plate System

Everything that breaks is something that moves

If you read the last post, you know the swappable plates now use screw-attachable magnets instead of glued ones. That sounds like a small change. Swap a bonded part for a fastened one, done.

It was not a small change, and chasing it properly is a large part of where the last ten months went. It also taught me where else to look, because once you understand why the magnets failed, you start noticing that the same thing is true elsewhere in the product.

The plates themselves came through it better than I expected. The magnets were made specifically for this design, so the plates kept their original outline and the same M3 holes they already had. What changed was the job those holes are doing. The plate stopped being a surface that holds switches and became a structural part, with the pull-out load of every swap running through it.

Which moved everything under it

Once the plate changed, everything it touches had to move with it. The plates sit at a fixed angle so they meet the pogo pin board correctly, and that geometry only works if every layer below is exactly where it is supposed to be. Change the thickness or stiffness of the top and you are re-checking the entire stack.

One result of that rework is a dedicated plate between the interface layer and the electronics, which also carries the pogo pin connector. Locating the connector off a single rigid part instead of off a chain of components means alignment stops depending on tolerances stacking up in your favour and starts being handled by the geometry itself.

The same problem, somewhere else entirely

Here is where the magnet failure earned its keep.

The lesson from the magnets was not really about magnets. It was that in a product like this, the parts that fail are the parts that get moved, and they usually fail at whatever is holding them rather than at the part itself. The magnet was fine. The bond was not.

So I went looking for anything else that gets moved repeatedly, and the obvious one was staring at me: the USB-C port.

A USB-C connector on a desk device gets plugged and unplugged thousands of times over the life of the product, and every one of those is a small lever action against the board it is soldered to. If the board can flex even slightly, that flex concentrates at the solder joints and the through-hole anchors of the connector. This is one of the most common ways consumer hardware dies, and it almost never dies because the connector wore out. It dies because the board moved.

The fix was structural pillars supporting the mainboard that carries the USB-C port and the storage, so the board is properly braced rather than held at its edges. Push a cable into the port now and the load goes into the enclosure instead of into the board.

It is an unglamorous change. It costs a little internal volume, adds features to the enclosure, and no customer will ever see it or thank me for it. It is also the difference between a product that lasts five years and one that develops an intermittent charging problem in year two.

And then the electronics

Once you open the mechanical design this far, the board follows.

The mainboard underwent design changes at both the routing and bill of materials level. Several passive components were removed because they provided minimal filtering benefit and negligible advantage to the USB lines, which operate at USB 2.0 speeds and therefore do not require additional protection. Additionally, the eMMC footprint was modified to accommodate a wider variety of different eMMC memory packages, mitigating supply chain risks from individual manufacturers. The most subtle change involved reducing the board from four layers to two, which decreased manufacturing costs by 3–4%.

Where things are now

The units on my desk have model numbers, CE marking and a UL file number on the bottom. That is a different kind of object from the one I was photographing a year ago. It is not a design I am still exploring, it is a design that has to be built repeatably by people who are not me.

Why the gap

I did not post for ten months because for most of that time the honest update would have been "still fixing it." I would rather publish one post that explains what actually changed than twelve saying progress is being made.

There is also a third thread I have barely touched here. A very large share of the last ten months went into the app rather than the hardware, and that deserves its own post. It is the next one.